Google AI Research Releases Agile Deliberation Framework for Gray-Area Image Classifiers
The framework helps users define subjective concepts like 'healthy food' by iteratively refining definitions using human feedback on borderline images.
Google AI Research has released Agile Deliberation, a human-in-the-loop framework for building image classifiers for subjective concepts. It uses prompts to Gemini models and retrieval with CLIP embeddings. The framework breaks concepts into unit concepts, finds borderline images, and iterates on user labels to optimize the definition. Evaluations show improved F1 scores over zero-shot approaches.
- Prompts Gemini 2.5 Pro and Flash without fine-tuning
- Uses CLIP embeddings and FAISS index
- 10.5 F1 points above zero-shot on average
- Demo notebook available
- Apache 2.0 licensed
- Vendors
- Products
- CLIP
- Organizations
- Google AI Research
- AI models
- Gemini 2.5 ProGemini 2.5 Flash
From the original article
7 min read at marktechpost.comGoogle Research released Agile Deliberation, a human-in-the-loop framework that helps users define subjective visual concepts through borderline images. In 18 live sessions, it lifted Gemini-based classifier F1 by 10. 5 points over zero-shot.
By Asif Razzaq · marktechpost.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.